Thematic analysis - How many codes do you need?

The safest way to start thematic analysis is to code broadly and in detail—even if the coding framework looks messy—because early-stage coding is meant to capture everything participants said before themes are formed. When codes are plentiful, researchers have a “backbone” that later supports refining, merging, and interpreting patterns; when codes are too few or too abstract, important evidence can be missed and later theme development becomes guesswork.

A common problem during coding is not having too much work, but having too little coverage. Detailed coding functions like a table of contents for the dataset: each code is a navigational entry that ensures transcripts don’t need to be revisited repeatedly. The workflow described here emphasizes a shift from reading data as individual stories to viewing it through a structured set of codes that represent the whole dataset holistically. That structure matters because themes don’t simply “emerge” on their own; researchers typically decide which themes to develop based on evidence, and that evidence can only be reliably organized through extensive coding.

The argument for “many codes” rests on three interconnected reasons. First, comprehensive coding prevents loss of content. If researchers code only lightly, they risk leaving parts of interviews uncovered, which undermines confidence that nothing important was overlooked. Second, detailed descriptive coding helps researchers avoid premature judgments about relevance. At the start, it’s impossible to know what will later prove important for the research questions, so the coding stage should capture experiences in concrete terms rather than prematurely labeling them with broad, interpretive concepts.

An example illustrates the difference between descriptive and overly abstract coding. If a participant describes being bullied and then reframing it as something that makes them stronger, a descriptive approach would code the bullying experience and the coping strategy (e.g., “turning challenges into advantages”) rather than collapsing it immediately into a high-level idea like “identity transformation.” That level of specificity creates multiple entry points for later analysis. When similar coping strategies appear elsewhere, the researcher can recognize a recurring pattern and legitimately elevate it into a theme or subtheme.

Third, extensive descriptive coding supports validity by reducing researcher bias. The early coding stage is treated as a safeguard against letting prior knowledge or theoretical expectations dictate labels too soon. By summarizing what participants said—rather than interpreting what they “must mean”—researchers delay abstraction until there is enough coded evidence to justify it. Over time, codes can be transformed into more interpretive, higher-level themes, but that interpretive step comes after the dataset has been thoroughly organized.

Overall, the guidance is pragmatic: if a coding framework feels overwhelming, that’s often a sign of adequate coverage. The recommendation is to keep many descriptive codes at first, then systematically reduce and refine them later—because starting with too few codes can leave researchers without the evidence base needed for credible, research-question-relevant themes.